Project Summary:

This project tackles the challenge of real-time music improvisation, particularly focusing on drum accompaniment, with the aim of generating drum tracks that closely resemble human performance. Our innovative approach involves leveraging a Conditional Generative Adversarial Network (CGAN) and Transformer models to enable a drummer AI to dynamically respond to a human bass guitarist during live performances. Unlike conventional methods, our emphasis lies in enhancing the responsiveness of generated music to human bassists and learning from human drummers by incorporating pitch velocity information. The CGAN framework facilitates the adaptation of accompaniment to the evolving bassline, fostering a seamless and spontaneous musical experience, while the Transformer model assigns velocity to the generated drum tracks, thereby enriching expressiveness. Furthermore, the CGAN framework allows for the selection of genre for the generated drum track among twelve different genres. The AI architecture is designed to enable real-time interaction with users for collaborative jamming sessions. This approach explores the accompaniment of humans with various instruments and endeavors to make generated music sound more human-like by learning from human players.

Created by Arash Sadeghi, Andrew Vardy, Andrew Staniland  @ MEARL

funded by IICSI / SSHRC Improvising Futures